Ecosyste.ms: Timeline
Browse the timeline of events for every public repo on GitHub. Data updated hourly from GH Archive.
zRzRzRzRzRzRzR created a comment on an issue on THUDM/CogVideo
Please read the readme carefully. I'm not sure if you changed your frame count to 81, and I noticed that you only used 20 steps instead of the official 50 steps, which has a significant impact on t...
zRzRzRzRzRzRzR pushed 4 commits to CogVideoX_dev THUDM/CogVideo
- [fix]fix deepspeed initialization issue in finetune examples ac2f2c7
- Merge pull request #565 from THUDM/CogVideoX_dev Cog video x dev a094b34
- Merge pull request #515 from Gforky/fix_finetune_demo [fix]fix deepspeed initialization issue in finetune examples 5aa6d3a
- Merge pull request #567 from THUDM/main New Finetune 87ccd38
zRzRzRzRzRzRzR pushed 2 commits to main THUDM/CogVideo
zRzRzRzRzRzRzR closed a pull request on THUDM/CogVideo
[fix]fix deepspeed initialization issue in finetune examples
修复 finetune 文件夹中 lora 微调脚本中deepspeed 初始化问题，目前脚本如果使用 deepspeed 会遇到 scheduler 初始化问题。
zRzRzRzRzRzRzR pushed 24 commits to main zRzRzRzRzRzRzR/transformers
- [`FlexAttention`] Update gemma2 (#34942) * update tests * now maybe this fixes the previous fialing tests! * nit d... 4c1388f
- Fix : Add PEFT from source to CI docker (#34969) * Docker fix peft * Test new docker * uncomment 8f48ccf
- Avoid calling `get_max_length` (#34971) fix Co-authored-by: ydshieh <[email protected]> 0d99a93
- Fix flaky test execution caused by `Thread` (#34966) fix Co-authored-by: ydshieh <[email protected]> 5f8b24e
- 🌐 [i18n-KO] Translated encoder-decoder.md to Korean (#34880) * Initial version of translation, english still remaini... 0600f46
- [docs] add explanation to `release_memory()` (#34911) * explain release_memory * Update docs/source/en/llm_tutori... 6372255
- [i18n-zh]Translated perf_train_special.md into Chinese (#34948) * Add translation for perf_train_special documentati... 2910015
- Fix typo in code block in vipllava.md (#34957) fix typo in code block in vipllava.md 4120cb2
- Fixed typo in `VisitWebpageTool` (#34978) Fixed typo in VisitWebpageTool 5523e38
- [PEFT] Set eval mode when loading PEFT adapter (#34509) * [PEFT] Set eval mode when loading PEFT adapter Resolves #... f4b674f
- Fix `save_pretrained` for partially offloaded models (#34890) * delete unnecessary reference Signed-off-by: Kyle Sa... 4f0bf98
- 🚨🚨🚨 Changed DINOv2Config default patch size to 14 (#34568) Changed DINOv2Config default patch size to 14 2b053fd
- Refine the code of Universal Assisted Generation (#34823) * removed the useless attritbutes * add configs for windo... 44af935
- Allow compressed-tensors quantized model to be trained (#34520) * populate quantization_config for kv-cache-scheme o... 57ca9e6
- Offloaded cache: fix generate (#34921) * fix cache impl * require_torch_gpu * fix mamba * fix copies 5e8c1d7
- Fix `utils/check_bad_commit.py` (for auto ping in CI) (#34943) * fix * fix --------- Co-authored-by: ydshie... 6300212
- Add optimized `PixtralImageProcessorFast` (#34836) * Add optimized PixtralImageProcessorFast * make style * Add du... 9d6f0dd
- Improve `.from_pretrained` type annotations (#34973) * Fix from_pretrained type annotations * Better typing for i... 01ad80f
- Fix docker CI : install autogptq from source (#35000) * Fixed Docker * Test ci * Finally * add comment f491096
- Let server decide default repo visibility (#34999) * Let server decide default repo visibility * code style 0b5b5e6
- and 4 more ...
zRzRzRzRzRzRzR pushed 17 commits to main zRzRzRzRzRzRzR/diffusers
- [Tests] skip nan lora tests on PyTorch 2.5.1 CPU. (#9975) * skip nan lora tests on PyTorch 2.5.1 CPU. * cog * ... 2e86a3f
- make `pipelines` tests device-agnostic (part1) (#9399) * enable on xpu * add 1 more * add one more * enable... 64b3e0f
- ControlNet from_single_file when already converted (#9978) Co-authored-by: Dhruv Nair <[email protected]> b5fd6f1
- Flux Fill, Canny, Depth, Redux (#9985) * update --------- Co-authored-by: yiyixuxu <[email protected]> Co-aut... 7ac6e28
- [SD3 dreambooth lora] smol fix to checkpoint saving (#9993) * smol change to fix checkpoint saving & resuming (as do... c4b5d2f
- [Docs] add: missing pipelines from the spec. (#10005) add: missing pipelines from the spec. 047bf49
- Add prompt about wandb in examples/dreambooth/readme. (#10014) Add files via upload 074e123
- [docs] Fix CogVideoX table (#10008) * fix * fix ad5ecd1
- Notebooks for Community Scripts-3 (#10032) * Add Notebooks for Community Scripts in ReadME. * Minor Script Impr... 8d477da
- Sd35 controlnet (#10020) * add model/pipeline Co-authored-by: Sayak Paul <[email protected]> 75bd1e8
- Add `beta`, `exponential` and `karras` sigmas to `FlowMatchEulerDiscreteScheduler` (#10001) Add beta, exponential an... e47cc1f
- Update sdxl reference pipeline to latest sdxl pipeline (#9938) * Update sdxl reference community pipeline * Updat... e44fc75
- [Community Pipeline] Add some feature for regional prompting pipeline (#9874) * [Fix] fix bugs of regional_promptin... 69c83d6
- Add sdxl controlnet reference community pipeline (#9893) * Add reference_attn & reference_adain support for sdxl wit... 069186f
- Change image_gen_aux repository URL (#10048) change image_gen_aux repo url fdec8bd
- make `pipelines` tests device-agnostic (part2) (#9400) * enable on xpu * add 1 more * add one more * enable... 6b288ec
- [Mochi-1] ensuring to compute the fourier features in FP32 in Mochi encoder (#10031) compute fourier features in FP32. c96bfa5
zRzRzRzRzRzRzR created a comment on an issue on THUDM/CogVideo
这个模型不太支持FP16，因此在readme中，我们强烈推荐使用BF16，FP32肯定没问题，如果显存够并且能接受更慢的时间
zRzRzRzRzRzRzR closed an issue on THUDM/CogVideo
在V100下使用FP16推理只得到全黑图像
### System Info / 系統信息 显卡：V100-32GB，其余环境按requirements.txt安装，diffusers库从源码安装 ### Information / 问题信息 - [ ] The official example scripts / 官方的示例脚本 - [X] My own modified scripts / 我自己修改的脚本和任务 ### R...
zRzRzRzRzRzRzR closed an issue on THUDM/CogVideo
CogVideoX1.5-5B-I2V 生成图片效果和网页差别很大
我这边使用A100 80g diffusers= 0.32.0.dev0  生成的结果 https://github.com/user-attachments/assets...
zRzRzRzRzRzRzR pushed 3 commits to main THUDM/CogVideo
zRzRzRzRzRzRzR created a comment on an issue on THUDM/GLM-Edge
I roughly understand what you mean. The EOS token in the training of the GLM model is <|user|>, which indicates the start of the next model's role, rather than a role like <|end|>. This also excee...
zRzRzRzRzRzRzR created a comment on an issue on THUDM/CogVideo
我想知道网页版的是哪个？ 1. 这个分辨率确实太小了，我们的readme表格表明了分辨率的范围 2. 与 pipe是否在cuda无关
zRzRzRzRzRzRzR created a comment on an issue on THUDM/GLM-Edge
这部分由于我无法拿到对应的来自高通的开源工具包，因此，我们也没办法将对应属于高通的工具包进行开源
zRzRzRzRzRzRzR created a comment on an issue on THUDM/CogVideo
你用的是哪个版本微调的，CogVdieoX-Factory还是本仓库的lora，本仓库lora没维护，在用Factory做维护
zRzRzRzRzRzRzR closed an issue on THUDM/CogVideo
Lora training resources usage
### Feature request / 功能建议 Even if it costs more time, Is it possible to reduce the resource requirements to train Loras locally ? ### Motivation / 动机 It would help the community in L...
zRzRzRzRzRzRzR closed an issue on THUDM/CogVideo
Finetune stuck at vae encode
### System Info / 系統信息 11.8 pytorch 2.5.0 ### Information / 问题信息 - [ ] The official example scripts / 官方的示例脚本 - [ ] My own modified scripts / 我自己修改的脚本和任务 ### Reproduction / 复现过程 def encode...
zRzRzRzRzRzRzR closed an issue on THUDM/CogVideo
Question about the VAE upsampling
Hi, I am trying to understand the logic of the CogVideoXUpsample3D. I found that it seems like, for tensors that have an odd `t` dimension, the first frame will be treated separately for spatial on...
zRzRzRzRzRzRzR closed an issue on THUDM/CogVideo
Pylint alerts corrections as part of intervention experiment
### Feature request / 功能建议 I'd like to conduct a software engineering experiment regarding the benefit of Pylint alerts removal.The experiment is described [here](https://github.com/evidencebp/py...
zRzRzRzRzRzRzR closed an issue on THUDM/CogVideo
Training Detail of 3DVAE
### Feature request / 功能建议 Training code of 3DVAE is needed. The 3DVAE training detail is not listed in the paper, can you provide 3DVAE training code or training detail? including - data you ...